Understanding Generation Parameters

Beyond your text prompt and input images/layers, Fater's AI models offer various Parameters that allow you to fine-tune the generation process and influence the characteristics of the output. These settings are found in the Left Sidebar after you've selected an AI model.

While some parameters are common across many models, others are specific to certain AI capabilities (like image editing, upscaling, or video generation).


Common & Key Parameters

Here's an overview of some important parameters you'll frequently encounter:

  • Steps (Diffusion Quality/Detail):

    • Found in: Many image generation and upscaling models.

    • What it does: Controls the number of processing steps the AI takes to refine the output. Generally, more steps can lead to higher detail and coherence but also increase generation time. Fewer steps are faster but might produce less refined results.

    • Typical Range: Varies by model (e.g., 15-50 for some generation models, 4-100 for some upscalers).

  • Guidance Scale (Prompt Adherence):

    • Found in: Many image and video generation models.

    • What it does: Determines how strictly the AI should follow your text prompt.

      • Lower values: Give the AI more creative freedom, potentially deviating from the prompt for more artistic or unexpected results.

      • Higher values: Force the AI to adhere more closely to your prompt's details.

    • Typical Range: Often 0-100 or 0-1, depending on the model.

  • Seed:

    • Found in: Almost all AI generation models.

    • What it does: Controls the initial randomness. Locking the seed helps reproduce results or iterate with minor changes. (See Seeds for a full explanation).


Parameters for Image Editing & Inpainting

When using models in the "Edit" category (e.g., for inpainting or modifying masked areas):

  • Style Type:

    • What it does: Allows you to select a predefined artistic style (e.g., Realistic, Design, Anime) to influence the look of the generated content within the masked area. "AUTO" lets the AI decide.

  • Destroy Fill Area Behind Mask:

    • What it does: A boolean (Yes/No) option. If "Yes," the AI completely ignores the original image content under your mask. If "No," the existing pixels might influence the new generation.

  • Mask Alpha Radius (Edge Blending):

    • What it does: Determines the smoothness of the transition between the AI-generated content and the original image at the edges of your mask. Higher values create a softer, more feathered blend.

  • Mask Extra Radius (Generation Padding):

    • What it does: The AI generates content slightly beyond your defined mask by this many pixels. This can help create a more seamless integration by providing overlap for blending.


Parameters for Upscaling

When using models in the "Upscale" category:

  • Scale vs. Output Megapixels:

    • Scale: A direct multiplier for the image dimensions (e.g., 2x, 4x).

    • Output Megapixels: Allows you to target a specific final resolution in millions of pixels.

  • Creativity:

    • What it does: Controls how much new detail the AI is allowed to invent during the upscaling process. Higher values can add more texture or fine details but might also introduce artifacts if pushed too far.

  • Resemblance:

    • What it does: Determines how faithfully the upscaled image should stick to the forms and structures of the original low-resolution image. Higher values mean less deviation.

  • Sharpen:

    • What it does: Applies an additional sharpening effect to the upscaled result.

  • Dynamic (HDR):

    • What it does: Influences the High Dynamic Range of the output, potentially enhancing contrast and lighting.

  • Tile Size:

    • What it does: An internal processing parameter. Smaller tiles might be faster or use less memory but could lead to visible seams in some cases. Larger tiles might produce more coherent results but be slower.

  • More Details:

    • What it does: An explicit control to encourage the AI to add finer details during upscaling.

  • Hand Fix:

    • What it does: A boolean (Enabled/Disabled) option to activate a specialized module aimed at improving the rendering of hands and fingers in upscaled images.

  • Checkpoint:

    • What it does: Allows selection of a specific underlying base model or training version that the upscaler uses, which can affect the style and quality of the output.


Parameters for Video Generation

When using models in the Video Editor:

  • Duration:

    • What it does: Sets the length of the generated video clip in seconds.


Finding and Understanding Parameters:

  • Always check the Left Sidebar after selecting an AI model to see its available parameters.

  • Many parameters include a brief description below the control to explain their function.

  • The AI Model Directory provides a more detailed breakdown of parameters for each user-facing model.

  • Experiment! The best way to understand how parameters affect your results is to try different settings. Remember to use the Restore Defaults button if you want to reset to the model's standard configuration.

Adjusting these parameters gives you significant creative control over the AI's output in Fater.

Last updated